GPT-trainer is the first and most powerful no-code AI Chat framework that implements a multi-agent architecture. Multiple AI Agents work together in a coordinated orchestration to enable advanced workflow automations.

We present this guide to help you understand how our Agents work together, as well as share with you some best practices surrounding multi-agent chatbot design.

All Agents have two states:

  • Active: Agent is “Connected”. It is live and running.
  • Inactive: Agent is “Disabled”. It does not do anything.

Only a single Agent engages with the user at any given time. Our AI Supervisor works in the background, in real time, to decide which Agent processes the user’s inquiry when it comes in.

Agents are completely isolated from one another. This means that you cannot tell Agent B what to do from inside Agent A, and you cannot tell the AI Supervisor to “pass the baton” from Agent A to Agent B directly. The AI Supervisor cannot be instructed on what to do using natural language.

The AI Supervisor assigns user queries to AI Agents based on the Agent description. This means that if you have only a single AI Agent, then you don’t need a very good description. But if you are working with multiple AI Agents, then you should have as explict a description as possible.

  • Agent Description - What kind of tasks should the Agent do?
  • Agent Prompt - How should the Agent perform its tasks?

AI Agent description is different from the Prompt. Description should tell the AI Supervisor what the Agent is designed to do at a high level, focusing on the type of user queries it should handle. Prompt is a set of instructions telling the AI Agent how to behave.

Good Agent descriptions should answer the question “What type of queries should the Agent handle?” More specifically, Agent descriptions should encompass all relevant “user intents” that it is supposed to handle.

Agent descriptions should not include details about “how the Agent performs its job”. That part should be reserved for the Agent’s Prompt.

Here are some examples of good Agent descriptions:

So what makes for a good multi-agent chatbot design? How can I ensure that all my user intents are handled by the appropriate Agent?

In general, a multi-agent architecture is great for having specialized AI Agents work together as a team. Each AI Agent can be trained on a different set of source data, be powered by a different LLM, have a different Prompt, or even be given different tools (functions). It is much more effective than dumping everything into a single Agent.

On the surface, your users will not know that they are engaging with multiple AI Agents when talking to the chatbot, since all of the complex orchestration takes place seamlessly behind the scenes. With the right setup, they will feel as though they are talking to a human instead of a machine. A multi-agent chatbot has the potential to be much more powerful than single-agent ones like ChatGPT.

In our AI Supervisor routing design, we use an algorithm that balances latency with performance. It will not always give you 100% accuracy, but you can get pretty close if you design your chatbots effectively.

​
Debugging the chatbot

If you notice that the chatbot is not responding to your expectations, then there may be two causes.

  1. The wrong Agent may have been assigned to handle the incoming user query.
  2. The Agent itself may not be optimally configured.

To find out whether your intended Agent has been selected to handle the user’s query, go to “Inbox” and turn on debug mode. Toggle it on here:

Under “Active Agent”, you will see the Agent that generated the response. Please note that you will only see “Active Agent” metadata if you have more than one connected AI Agents.

If the Agent that handled the user query was not the correct one, you can return to Agent configuration and update its name, description, or prompt.

If the correct Agent handled the user query but the response was not what you expected, then the issue resides within the Agent itself. You will then need to update the Agent configuration. The most straightforward way to fix a response is by revising the AI’s answer directly:

This ensures that the chatbot will generate an answer based on what you revised every time it encounters the question in the future, assuming the same Agent was chosen to respond to the query.

If you’d like to make the Agent more robust and consistent in general, you can curate the training data, choose a more powerful LLM, or be more explicit in your base prompt.

To learn some best practices for training the chatbot, see best practices for preparing training data

To learn more about fine-tuning Agent intents, please check out the article Fine-Tuning Agent Intents.

Conversation LabelingFine Tuning Agent Intents